Some appMalloc optimizations

- CTextureData::Decompress using appMalloc(noinit)
- indicating init/noinit flag for appMalloc in profiler

Other: optimized DXT texture decompression
- NVTT: using "optimize by speed" mode
- common.project: using "fast floating point" mode for Visual C++ (-fp:fast), otherwise NVTT's "optimize by speed" was failed to locate modern __libm_sse2_sqrt_precise function (UCRT function)

Core/Memory.cpp

@@ -130,6 +130,7 @@ inline void OutOfMemory(int size)
 void* appMalloc(int size, int alignment, bool noInit)
 {
 	guard(appMalloc);
+	PROFILE_LABEL(noInit ? "NoInit" : "Zero");
 
 #if DEBUG_MEMORY
 	// Reserve some amount of memory for possibility to log memory when crashed

Unreal/UnrealMaterial/UnTexture.cpp

@@ -113,7 +113,7 @@ byte* CTextureData::Decompress(int MipLevel)
 
 	int pixelSize = PixelFormatInfo[Format].Float ? 16 : 4;
 	int size = USize * VSize * pixelSize;
-	byte* dst = new byte [size];
+	byte* dst = (byte*)appMalloc(size, DEFAULT_ALIGNMENT, true);
 
 #if 0
 	{

common.project

@@ -40,6 +40,7 @@ USE_SYSTEM_LIBS = 0
 		STDDEFS += _NO_CRT_STDIO_INLINE					# UCRT (VS 2015+): without that, all printf functions will use some "common" function
 		!include $WDKCRT/msvcrt.project
 		!if "$VC_VER" >= 2015							# support for OLDCRT with modern compliers
+			OPTIONS += -fp:fast
 			OPTIONS += -Zc:threadSafeInit-				# disable calls to C++11 thread-safe static object initializers
 			STDDEFS += _ACRTIMP_ALT=					# remove dllimport from some CRT functions (there're just a few)
 		!endif
@@ -93,8 +94,10 @@ OBJDIR = $R/obj/libs-$PLATFORM
 !endif
 
 push(INCLUDES)
+push(OPTIMIZE)
 INCLUDES = $R/libs/nvtt
 LIBINCLUDES += $R/libs/nvtt
+OPTIMIZE = speed
 
 sources(NV_LIBS) = {
 #	$R/libs/nvtt/nvcore/*.cpp
@@ -104,6 +107,7 @@ sources(NV_LIBS) = {
 #	$R/libs/nvtt/nvtt/bc7/*.cpp
 }
 
+pop(OPTIMIZE)
 pop(INCLUDES)